@ccie4526 I am about to board a plane but three quick tips:
1. Make agents with personalities (e.g. look into Soul.md)
2. Build your own skills or get ones you can relatively trust from ClawHub.
Those two will help guide agents a lot. Think of them as people, maybe like hiring a contractor that can do specific tasks.
Start with a high-level objective. Then just ask how to best accomplish it. Go from a plan (either in a planning mode or just interactively) to an implementation process. Then have a deploy and a monitor process.
Breaking down work like that will get you going.
I would strongly suggest starting on something that is a reasonable size that you want to do, but have never had time or expertise to do. Use that passion project as a starting point. Do not do something too small because the models can do a lot. You want to get to the experience where you can walk away for a few hours and have it work.
3. When things break, it is often due to a small/ older model. That is not just in the actual workflow. Using something like an Opus/ Sonnet or Codex to set everything up is actually quite helpful.
Bonus tip: Get the agent a web browser it can use, and point it to research. For example, if you are configuring a MikroTik switch, tell it to look up documentation or point it to documentation you had bookmarked.
I almost wonder if this is worth having its own thread.
